src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <dirent.h>
   4 #include <errno.h>
   5 #include <fcntl.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <string.h>
   9 #include <sys/file.h>
  10 #include <sys/stat.h>
  11 #include <unistd.h>
  12 #include <linux/fs.h>
  13
  14 #include "alloc-util.h"
  15 #include "btrfs-util.h"
  16 #include "chattr-util.h"
  17 #include "copy.h"
  18 #include "dirent-util.h"
  19 #include "dissect-image.h"
  20 #include "env-file.h"
  21 #include "env-util.h"
  22 #include "fd-util.h"
  23 #include "fs-util.h"
  24 #include "hashmap.h"
  25 #include "hostname-util.h"
  26 #include "id128-util.h"
  27 #include "lockfile-util.h"
  28 #include "log.h"
  29 #include "loop-util.h"
  30 #include "machine-image.h"
  31 #include "macro.h"
  32 #include "mkdir.h"
  33 #include "os-util.h"
  34 #include "path-util.h"
  35 #include "rm-rf.h"
  36 #include "string-table.h"
  37 #include "string-util.h"
  38 #include "strv.h"
  39 #include "time-util.h"
  40 #include "utf8.h"
  41 #include "util.h"
  42 #include "xattr-util.h"
  43
  44 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  45         [IMAGE_MACHINE] =  "/etc/machines\0"              /* only place symlinks here */
  46                            "/run/machines\0"              /* and here too */
  47                            "/var/lib/machines\0"          /* the main place for images */
  48                            "/var/lib/container\0"         /* legacy */
  49                            "/usr/local/lib/machines\0"
  50                            "/usr/lib/machines\0",
  51
  52         [IMAGE_PORTABLE] = "/etc/portables\0"             /* only place symlinks here */
  53                            "/run/portables\0"             /* and here too */
  54                            "/var/lib/portables\0"         /* the main place for images */
  55                            "/usr/local/lib/portables\0"
  56                            "/usr/lib/portables\0",
  57 };
  58
  59 static Image *image_free(Image *i) {
  60         assert(i);
  61
  62         free(i->name);
  63         free(i->path);
  64
  65         free(i->hostname);
  66         strv_free(i->machine_info);
  67         strv_free(i->os_release);
  68
  69         return mfree(i);
  70 }
  71
  72 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  73 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  74                                       Image, image_unref);
  75
  76 static char **image_settings_path(Image *image) {
  77         _cleanup_strv_free_ char **l = NULL;
  78         const char *fn, *s;
  79         unsigned i = 0;
  80
  81         assert(image);
  82
  83         l = new0(char*, 4);
  84         if (!l)
  85                 return NULL;
  86
  87         fn = strjoina(image->name, ".nspawn");
  88
  89         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
  90                 l[i] = strappend(s, fn);
  91                 if (!l[i])
  92                         return NULL;
  93
  94                 i++;
  95         }
  96
  97         l[i] = file_in_same_dir(image->path, fn);
  98         if (!l[i])
  99                 return NULL;
 100
 101         return TAKE_PTR(l);
 102 }
 103
 104 static char *image_roothash_path(Image *image) {
 105         const char *fn;
 106
 107         assert(image);
 108
 109         fn = strjoina(image->name, ".roothash");
 110
 111         return file_in_same_dir(image->path, fn);
 112 }
 113
 114 static int image_new(
 115                 ImageType t,
 116                 const char *pretty,
 117                 const char *path,
 118                 const char *filename,
 119                 bool read_only,
 120                 usec_t crtime,
 121                 usec_t mtime,
 122                 Image **ret) {
 123
 124         _cleanup_(image_unrefp) Image *i = NULL;
 125
 126         assert(t >= 0);
 127         assert(t < _IMAGE_TYPE_MAX);
 128         assert(pretty);
 129         assert(filename);
 130         assert(ret);
 131
 132         i = new0(Image, 1);
 133         if (!i)
 134                 return -ENOMEM;
 135
 136         i->n_ref = 1;
 137         i->type = t;
 138         i->read_only = read_only;
 139         i->crtime = crtime;
 140         i->mtime = mtime;
 141         i->usage = i->usage_exclusive = (uint64_t) -1;
 142         i->limit = i->limit_exclusive = (uint64_t) -1;
 143
 144         i->name = strdup(pretty);
 145         if (!i->name)
 146                 return -ENOMEM;
 147
 148         if (path)
 149                 i->path = strjoin(path, "/", filename);
 150         else
 151                 i->path = strdup(filename);
 152         if (!i->path)
 153                 return -ENOMEM;
 154
 155         path_simplify(i->path, false);
 156
 157         *ret = TAKE_PTR(i);
 158
 159         return 0;
 160 }
 161
 162 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 163         _cleanup_free_ char *name = NULL;
 164         const char *p;
 165         size_t n;
 166
 167         assert(path);
 168         assert(ret);
 169
 170         p = last_path_component(path);
 171         n = strcspn(p, "/");
 172
 173         name = strndup(p, n);
 174         if (!name)
 175                 return -ENOMEM;
 176
 177         if (suffix) {
 178                 char *e;
 179
 180                 e = endswith(name, suffix);
 181                 if (!e)
 182                         return -EINVAL;
 183
 184                 *e = 0;
 185         }
 186
 187         if (!image_name_is_valid(name))
 188                 return -EINVAL;
 189
 190         *ret = TAKE_PTR(name);
 191         return 0;
 192 }
 193
 194 static int image_make(
 195                 const char *pretty,
 196                 int dfd,
 197                 const char *path,
 198                 const char *filename,
 199                 const struct stat *st,
 200                 Image **ret) {
 201
 202         _cleanup_free_ char *pretty_buffer = NULL;
 203         struct stat stbuf;
 204         bool read_only;
 205         int r;
 206
 207         assert(dfd >= 0 || dfd == AT_FDCWD);
 208         assert(filename);
 209
 210         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 211          * devices into /var/lib/machines/, and treat them normally.
 212          *
 213          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 214          * recognize. */
 215
 216         if (!st) {
 217                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 218                         return -errno;
 219
 220                 st = &stbuf;
 221         }
 222
 223         read_only =
 224                 (path && path_startswith(path, "/usr")) ||
 225                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 226
 227         if (S_ISDIR(st->st_mode)) {
 228                 _cleanup_close_ int fd = -1;
 229                 unsigned file_attr = 0;
 230
 231                 if (!ret)
 232                         return 0;
 233
 234                 if (!pretty) {
 235                         r = extract_pretty(filename, NULL, &pretty_buffer);
 236                         if (r < 0)
 237                                 return r;
 238
 239                         pretty = pretty_buffer;
 240                 }
 241
 242                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 243                 if (fd < 0)
 244                         return -errno;
 245
 246                 /* btrfs subvolumes have inode 256 */
 247                 if (st->st_ino == 256) {
 248
 249                         r = btrfs_is_filesystem(fd);
 250                         if (r < 0)
 251                                 return r;
 252                         if (r) {
 253                                 BtrfsSubvolInfo info;
 254
 255                                 /* It's a btrfs subvolume */
 256
 257                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 258                                 if (r < 0)
 259                                         return r;
 260
 261                                 r = image_new(IMAGE_SUBVOLUME,
 262                                               pretty,
 263                                               path,
 264                                               filename,
 265                                               info.read_only || read_only,
 266                                               info.otime,
 267                                               0,
 268                                               ret);
 269                                 if (r < 0)
 270                                         return r;
 271
 272                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 273                                         BtrfsQuotaInfo quota;
 274
 275                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 276                                         if (r >= 0) {
 277                                                 (*ret)->usage = quota.referenced;
 278                                                 (*ret)->usage_exclusive = quota.exclusive;
 279
 280                                                 (*ret)->limit = quota.referenced_max;
 281                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 282                                         }
 283                                 }
 284
 285                                 return 0;
 286                         }
 287                 }
 288
 289                 /* If the IMMUTABLE bit is set, we consider the
 290                  * directory read-only. Since the ioctl is not
 291                  * supported everywhere we ignore failures. */
 292                 (void) read_attr_fd(fd, &file_attr);
 293
 294                 /* It's just a normal directory. */
 295                 r = image_new(IMAGE_DIRECTORY,
 296                               pretty,
 297                               path,
 298                               filename,
 299                               read_only || (file_attr & FS_IMMUTABLE_FL),
 300                               0,
 301                               0,
 302                               ret);
 303                 if (r < 0)
 304                         return r;
 305
 306                 return 0;
 307
 308         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 309                 usec_t crtime = 0;
 310
 311                 /* It's a RAW disk image */
 312
 313                 if (!ret)
 314                         return 0;
 315
 316                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 317
 318                 if (!pretty) {
 319                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 320                         if (r < 0)
 321                                 return r;
 322
 323                         pretty = pretty_buffer;
 324                 }
 325
 326                 r = image_new(IMAGE_RAW,
 327                               pretty,
 328                               path,
 329                               filename,
 330                               !(st->st_mode & 0222) || read_only,
 331                               crtime,
 332                               timespec_load(&st->st_mtim),
 333                               ret);
 334                 if (r < 0)
 335                         return r;
 336
 337                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 338                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 339
 340                 return 0;
 341
 342         } else if (S_ISBLK(st->st_mode)) {
 343                 _cleanup_close_ int block_fd = -1;
 344                 uint64_t size = UINT64_MAX;
 345
 346                 /* A block device */
 347
 348                 if (!ret)
 349                         return 0;
 350
 351                 if (!pretty) {
 352                         r = extract_pretty(filename, NULL, &pretty_buffer);
 353                         if (r < 0)
 354                                 return r;
 355
 356                         pretty = pretty_buffer;
 357                 }
 358
 359                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 360                 if (block_fd < 0)
 361                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 362                 else {
 363                         /* Refresh stat data after opening the node */
 364                         if (fstat(block_fd, &stbuf) < 0)
 365                                 return -errno;
 366                         st = &stbuf;
 367
 368                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 369                                 return -ENOTTY;
 370
 371                         if (!read_only) {
 372                                 int state = 0;
 373
 374                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 375                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 376                                 else if (state)
 377                                         read_only = true;
 378                         }
 379
 380                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 381                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path, filename);
 382
 383                         block_fd = safe_close(block_fd);
 384                 }
 385
 386                 r = image_new(IMAGE_BLOCK,
 387                               pretty,
 388                               path,
 389                               filename,
 390                               !(st->st_mode & 0222) || read_only,
 391                               0,
 392                               0,
 393                               ret);
 394                 if (r < 0)
 395                         return r;
 396
 397                 if (size != 0 && size != UINT64_MAX)
 398                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 399
 400                 return 0;
 401         }
 402
 403         return -EMEDIUMTYPE;
 404 }
 405
 406 int image_find(ImageClass class, const char *name, Image **ret) {
 407         const char *path;
 408         int r;
 409
 410         assert(class >= 0);
 411         assert(class < _IMAGE_CLASS_MAX);
 412         assert(name);
 413
 414         /* There are no images with invalid names */
 415         if (!image_name_is_valid(name))
 416                 return -ENOENT;
 417
 418         NULSTR_FOREACH(path, image_search_path[class]) {
 419                 _cleanup_closedir_ DIR *d = NULL;
 420                 struct stat st;
 421
 422                 d = opendir(path);
 423                 if (!d) {
 424                         if (errno == ENOENT)
 425                                 continue;
 426
 427                         return -errno;
 428                 }
 429
 430                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 431                  * symlink block devices into the search path */
 432                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 433                         _cleanup_free_ char *raw = NULL;
 434
 435                         if (errno != ENOENT)
 436                                 return -errno;
 437
 438                         raw = strappend(name, ".raw");
 439                         if (!raw)
 440                                 return -ENOMEM;
 441
 442                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 443
 444                                 if (errno == ENOENT)
 445                                         continue;
 446
 447                                 return -errno;
 448                         }
 449
 450                         if (!S_ISREG(st.st_mode))
 451                                 continue;
 452
 453                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 454
 455                 } else {
 456                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 457                                 continue;
 458
 459                         r = image_make(name, dirfd(d), path, name, &st, ret);
 460                 }
 461                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 462                         continue;
 463                 if (r < 0)
 464                         return r;
 465
 466                 if (ret)
 467                         (*ret)->discoverable = true;
 468
 469                 return 1;
 470         }
 471
 472         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 473                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 474                 if (r < 0)
 475                         return r;
 476
 477                 if (ret)
 478                         (*ret)->discoverable = true;
 479
 480                 return r;
 481         }
 482
 483         return -ENOENT;
 484 };
 485
 486 int image_from_path(const char *path, Image **ret) {
 487
 488         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 489          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 490          * overridden by another, different image earlier in the search path */
 491
 492         if (path_equal(path, "/"))
 493                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 494
 495         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 496 }
 497
 498 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 499         if (image_name_is_valid(name_or_path))
 500                 return image_find(class, name_or_path, ret);
 501
 502         return image_from_path(name_or_path, ret);
 503 }
 504
 505 int image_discover(ImageClass class, Hashmap *h) {
 506         const char *path;
 507         int r;
 508
 509         assert(class >= 0);
 510         assert(class < _IMAGE_CLASS_MAX);
 511         assert(h);
 512
 513         NULSTR_FOREACH(path, image_search_path[class]) {
 514                 _cleanup_closedir_ DIR *d = NULL;
 515                 struct dirent *de;
 516
 517                 d = opendir(path);
 518                 if (!d) {
 519                         if (errno == ENOENT)
 520                                 continue;
 521
 522                         return -errno;
 523                 }
 524
 525                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 526                         _cleanup_(image_unrefp) Image *image = NULL;
 527                         _cleanup_free_ char *truncated = NULL;
 528                         const char *pretty;
 529                         struct stat st;
 530
 531                         if (dot_or_dot_dot(de->d_name))
 532                                 continue;
 533
 534                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 535                          * to symlink block devices into the search path */
 536                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 537                                 if (errno == ENOENT)
 538                                         continue;
 539
 540                                 return -errno;
 541                         }
 542
 543                         if (S_ISREG(st.st_mode)) {
 544                                 const char *e;
 545
 546                                 e = endswith(de->d_name, ".raw");
 547                                 if (!e)
 548                                         continue;
 549
 550                                 truncated = strndup(de->d_name, e - de->d_name);
 551                                 if (!truncated)
 552                                         return -ENOMEM;
 553
 554                                 pretty = truncated;
 555                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 556                                 pretty = de->d_name;
 557                         else
 558                                 continue;
 559
 560                         if (!image_name_is_valid(pretty))
 561                                 continue;
 562
 563                         if (hashmap_contains(h, pretty))
 564                                 continue;
 565
 566                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 567                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 568                                 continue;
 569                         if (r < 0)
 570                                 return r;
 571
 572                         image->discoverable = true;
 573
 574                         r = hashmap_put(h, image->name, image);
 575                         if (r < 0)
 576                                 return r;
 577
 578                         image = NULL;
 579                 }
 580         }
 581
 582         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 583                 _cleanup_(image_unrefp) Image *image = NULL;
 584
 585                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 586                 if (r < 0)
 587                         return r;
 588
 589                 image->discoverable = true;
 590
 591                 r = hashmap_put(h, image->name, image);
 592                 if (r < 0)
 593                         return r;
 594
 595                 image = NULL;
 596         }
 597
 598         return 0;
 599 }
 600
 601 int image_remove(Image *i) {
 602         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 603         _cleanup_strv_free_ char **settings = NULL;
 604         _cleanup_free_ char *roothash = NULL;
 605         char **j;
 606         int r;
 607
 608         assert(i);
 609
 610         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 611                 return -EROFS;
 612
 613         settings = image_settings_path(i);
 614         if (!settings)
 615                 return -ENOMEM;
 616
 617         roothash = image_roothash_path(i);
 618         if (!roothash)
 619                 return -ENOMEM;
 620
 621         /* Make sure we don't interfere with a running nspawn */
 622         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 623         if (r < 0)
 624                 return r;
 625
 626         switch (i->type) {
 627
 628         case IMAGE_SUBVOLUME:
 629
 630                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 631                  * big guns */
 632                 if (unlink(i->path) < 0) {
 633                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 634                         if (r < 0)
 635                                 return r;
 636                 }
 637
 638                 break;
 639
 640         case IMAGE_DIRECTORY:
 641                 /* Allow deletion of read-only directories */
 642                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 643                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 644                 if (r < 0)
 645                         return r;
 646
 647                 break;
 648
 649         case IMAGE_BLOCK:
 650
 651                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 652                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 653                  * the thing (it's most likely a symlink after all). */
 654
 655                 if (path_startswith(i->path, "/dev"))
 656                         break;
 657
 658                 _fallthrough_;
 659         case IMAGE_RAW:
 660                 if (unlink(i->path) < 0)
 661                         return -errno;
 662                 break;
 663
 664         default:
 665                 return -EOPNOTSUPP;
 666         }
 667
 668         STRV_FOREACH(j, settings) {
 669                 if (unlink(*j) < 0 && errno != ENOENT)
 670                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 671         }
 672
 673         if (unlink(roothash) < 0 && errno != ENOENT)
 674                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 675
 676         return 0;
 677 }
 678
 679 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 680         _cleanup_free_ char *rs = NULL;
 681         const char *fn;
 682
 683         fn = strjoina(new_name, suffix);
 684
 685         rs = file_in_same_dir(path, fn);
 686         if (!rs)
 687                 return -ENOMEM;
 688
 689         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 690 }
 691
 692 int image_rename(Image *i, const char *new_name) {
 693         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 694         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 695         _cleanup_strv_free_ char **settings = NULL;
 696         unsigned file_attr = 0;
 697         char **j;
 698         int r;
 699
 700         assert(i);
 701
 702         if (!image_name_is_valid(new_name))
 703                 return -EINVAL;
 704
 705         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 706                 return -EROFS;
 707
 708         settings = image_settings_path(i);
 709         if (!settings)
 710                 return -ENOMEM;
 711
 712         roothash = image_roothash_path(i);
 713         if (!roothash)
 714                 return -ENOMEM;
 715
 716         /* Make sure we don't interfere with a running nspawn */
 717         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 718         if (r < 0)
 719                 return r;
 720
 721         /* Make sure nobody takes the new name, between the time we
 722          * checked it is currently unused in all search paths, and the
 723          * time we take possession of it */
 724         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 725         if (r < 0)
 726                 return r;
 727
 728         r = image_find(IMAGE_MACHINE, new_name, NULL);
 729         if (r >= 0)
 730                 return -EEXIST;
 731         if (r != -ENOENT)
 732                 return r;
 733
 734         switch (i->type) {
 735
 736         case IMAGE_DIRECTORY:
 737                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 738                 (void) read_attr_path(i->path, &file_attr);
 739
 740                 if (file_attr & FS_IMMUTABLE_FL)
 741                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 742
 743                 _fallthrough_;
 744         case IMAGE_SUBVOLUME:
 745                 new_path = file_in_same_dir(i->path, new_name);
 746                 break;
 747
 748         case IMAGE_BLOCK:
 749
 750                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 751                 if (path_startswith(i->path, "/dev"))
 752                         return -EROFS;
 753
 754                 new_path = file_in_same_dir(i->path, new_name);
 755                 break;
 756
 757         case IMAGE_RAW: {
 758                 const char *fn;
 759
 760                 fn = strjoina(new_name, ".raw");
 761                 new_path = file_in_same_dir(i->path, fn);
 762                 break;
 763         }
 764
 765         default:
 766                 return -EOPNOTSUPP;
 767         }
 768
 769         if (!new_path)
 770                 return -ENOMEM;
 771
 772         nn = strdup(new_name);
 773         if (!nn)
 774                 return -ENOMEM;
 775
 776         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 777         if (r < 0)
 778                 return r;
 779
 780         /* Restore the immutable bit, if it was set before */
 781         if (file_attr & FS_IMMUTABLE_FL)
 782                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 783
 784         free_and_replace(i->path, new_path);
 785         free_and_replace(i->name, nn);
 786
 787         STRV_FOREACH(j, settings) {
 788                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 789                 if (r < 0 && r != -ENOENT)
 790                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 791         }
 792
 793         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 794         if (r < 0 && r != -ENOENT)
 795                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 796
 797         return 0;
 798 }
 799
 800 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 801         _cleanup_free_ char *rs = NULL;
 802         const char *fn;
 803
 804         fn = strjoina(new_name, suffix);
 805
 806         rs = file_in_same_dir(path, fn);
 807         if (!rs)
 808                 return -ENOMEM;
 809
 810         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 811 }
 812
 813 int image_clone(Image *i, const char *new_name, bool read_only) {
 814         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 815         _cleanup_strv_free_ char **settings = NULL;
 816         _cleanup_free_ char *roothash = NULL;
 817         const char *new_path;
 818         char **j;
 819         int r;
 820
 821         assert(i);
 822
 823         if (!image_name_is_valid(new_name))
 824                 return -EINVAL;
 825
 826         settings = image_settings_path(i);
 827         if (!settings)
 828                 return -ENOMEM;
 829
 830         roothash = image_roothash_path(i);
 831         if (!roothash)
 832                 return -ENOMEM;
 833
 834         /* Make sure nobody takes the new name, between the time we
 835          * checked it is currently unused in all search paths, and the
 836          * time we take possession of it */
 837         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 838         if (r < 0)
 839                 return r;
 840
 841         r = image_find(IMAGE_MACHINE, new_name, NULL);
 842         if (r >= 0)
 843                 return -EEXIST;
 844         if (r != -ENOENT)
 845                 return r;
 846
 847         switch (i->type) {
 848
 849         case IMAGE_SUBVOLUME:
 850         case IMAGE_DIRECTORY:
 851                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 852                  * directory. */
 853
 854                 new_path = strjoina("/var/lib/machines/", new_name);
 855
 856                 r = btrfs_subvol_snapshot(i->path, new_path,
 857                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 858                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 859                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 860                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 861                                           BTRFS_SNAPSHOT_RECURSIVE |
 862                                           BTRFS_SNAPSHOT_QUOTA);
 863                 if (r >= 0)
 864                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 865                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 866
 867                 break;
 868
 869         case IMAGE_RAW:
 870                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 871
 872                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK);
 873                 break;
 874
 875         case IMAGE_BLOCK:
 876         default:
 877                 return -EOPNOTSUPP;
 878         }
 879
 880         if (r < 0)
 881                 return r;
 882
 883         STRV_FOREACH(j, settings) {
 884                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 885                 if (r < 0 && r != -ENOENT)
 886                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 887         }
 888
 889         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 890         if (r < 0 && r != -ENOENT)
 891                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 892
 893         return 0;
 894 }
 895
 896 int image_read_only(Image *i, bool b) {
 897         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 898         int r;
 899
 900         assert(i);
 901
 902         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 903                 return -EROFS;
 904
 905         /* Make sure we don't interfere with a running nspawn */
 906         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 907         if (r < 0)
 908                 return r;
 909
 910         switch (i->type) {
 911
 912         case IMAGE_SUBVOLUME:
 913
 914                 /* Note that we set the flag only on the top-level
 915                  * subvolume of the image. */
 916
 917                 r = btrfs_subvol_set_read_only(i->path, b);
 918                 if (r < 0)
 919                         return r;
 920
 921                 break;
 922
 923         case IMAGE_DIRECTORY:
 924                 /* For simple directory trees we cannot use the access
 925                    mode of the top-level directory, since it has an
 926                    effect on the container itself.  However, we can
 927                    use the "immutable" flag, to at least make the
 928                    top-level directory read-only. It's not as good as
 929                    a read-only subvolume, but at least something, and
 930                    we can read the value back. */
 931
 932                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 933                 if (r < 0)
 934                         return r;
 935
 936                 break;
 937
 938         case IMAGE_RAW: {
 939                 struct stat st;
 940
 941                 if (stat(i->path, &st) < 0)
 942                         return -errno;
 943
 944                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 945                         return -errno;
 946
 947                 /* If the images is now read-only, it's a good time to
 948                  * defrag it, given that no write patterns will
 949                  * fragment it again. */
 950                 if (b)
 951                         (void) btrfs_defrag(i->path);
 952                 break;
 953         }
 954
 955         case IMAGE_BLOCK: {
 956                 _cleanup_close_ int fd = -1;
 957                 struct stat st;
 958                 int state = b;
 959
 960                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 961                 if (fd < 0)
 962                         return -errno;
 963
 964                 if (fstat(fd, &st) < 0)
 965                         return -errno;
 966                 if (!S_ISBLK(st.st_mode))
 967                         return -ENOTTY;
 968
 969                 if (ioctl(fd, BLKROSET, &state) < 0)
 970                         return -errno;
 971
 972                 break;
 973         }
 974
 975         default:
 976                 return -EOPNOTSUPP;
 977         }
 978
 979         return 0;
 980 }
 981
 982 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 983         _cleanup_free_ char *p = NULL;
 984         LockFile t = LOCK_FILE_INIT;
 985         struct stat st;
 986         int r;
 987
 988         assert(path);
 989         assert(global);
 990         assert(local);
 991
 992         /* Locks an image path. This actually creates two locks: one
 993          * "local" one, next to the image path itself, which might be
 994          * shared via NFS. And another "global" one, in /run, that
 995          * uses the device/inode number. This has the benefit that we
 996          * can even lock a tree that is a mount point, correctly. */
 997
 998         if (!path_is_absolute(path))
 999                 return -EINVAL;
1000
1001         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1002                 *local = *global = (LockFile) LOCK_FILE_INIT;
1003                 return 0;
1004         }
1005
1006         if (path_equal(path, "/"))
1007                 return -EBUSY;
1008
1009         if (stat(path, &st) >= 0) {
1010                 if (S_ISBLK(st.st_mode))
1011                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1012                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1013                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1014                 else
1015                         return -ENOTTY;
1016
1017                 if (r < 0)
1018                         return -ENOMEM;
1019         }
1020
1021         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
1022          * block devices are device local anyway. */
1023         if (!path_startswith(path, "/dev")) {
1024                 r = make_lock_file_for(path, operation, &t);
1025                 if (r < 0) {
1026                         if ((operation & LOCK_SH) && r == -EROFS)
1027                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1028                         else
1029                                 return r;
1030                 }
1031         }
1032
1033         if (p) {
1034                 mkdir_p("/run/systemd/nspawn/locks", 0700);
1035
1036                 r = make_lock_file(p, operation, global);
1037                 if (r < 0) {
1038                         release_lock_file(&t);
1039                         return r;
1040                 }
1041         } else
1042                 *global = (LockFile) LOCK_FILE_INIT;
1043
1044         *local = t;
1045         return 0;
1046 }
1047
1048 int image_set_limit(Image *i, uint64_t referenced_max) {
1049         assert(i);
1050
1051         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1052                 return -EROFS;
1053
1054         if (i->type != IMAGE_SUBVOLUME)
1055                 return -EOPNOTSUPP;
1056
1057         /* We set the quota both for the subvolume as well as for the
1058          * subtree. The latter is mostly for historical reasons, since
1059          * we didn't use to have a concept of subtree quota, and hence
1060          * only modified the subvolume quota. */
1061
1062         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1063         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1064         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1065 }
1066
1067 int image_read_metadata(Image *i) {
1068         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1069         int r;
1070
1071         assert(i);
1072
1073         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1074         if (r < 0)
1075                 return r;
1076
1077         switch (i->type) {
1078
1079         case IMAGE_SUBVOLUME:
1080         case IMAGE_DIRECTORY: {
1081                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1082                 sd_id128_t machine_id = SD_ID128_NULL;
1083                 _cleanup_free_ char *hostname = NULL;
1084                 _cleanup_free_ char *path = NULL;
1085
1086                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1087                 if (r < 0 && r != -ENOENT)
1088                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1089                 else if (r >= 0) {
1090                         r = read_etc_hostname(path, &hostname);
1091                         if (r < 0)
1092                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1093                 }
1094
1095                 path = mfree(path);
1096
1097                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1098                 if (r < 0 && r != -ENOENT)
1099                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1100                 else if (r >= 0) {
1101                         _cleanup_close_ int fd = -1;
1102
1103                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1104                         if (fd < 0)
1105                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1106                         else {
1107                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1108                                 if (r < 0)
1109                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1110                         }
1111                 }
1112
1113                 path = mfree(path);
1114
1115                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1116                 if (r < 0 && r != -ENOENT)
1117                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1118                 else if (r >= 0) {
1119                         r = load_env_file_pairs(NULL, path, &machine_info);
1120                         if (r < 0)
1121                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1122                 }
1123
1124                 r = load_os_release_pairs(i->path, &os_release);
1125                 if (r < 0)
1126                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1127
1128                 free_and_replace(i->hostname, hostname);
1129                 i->machine_id = machine_id;
1130                 strv_free_and_replace(i->machine_info, machine_info);
1131                 strv_free_and_replace(i->os_release, os_release);
1132
1133                 break;
1134         }
1135
1136         case IMAGE_RAW:
1137         case IMAGE_BLOCK: {
1138                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1139                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1140
1141                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
1142                 if (r < 0)
1143                         return r;
1144
1145                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
1146                 if (r < 0)
1147                         return r;
1148
1149                 r = dissected_image_acquire_metadata(m);
1150                 if (r < 0)
1151                         return r;
1152
1153                 free_and_replace(i->hostname, m->hostname);
1154                 i->machine_id = m->machine_id;
1155                 strv_free_and_replace(i->machine_info, m->machine_info);
1156                 strv_free_and_replace(i->os_release, m->os_release);
1157
1158                 break;
1159         }
1160
1161         default:
1162                 return -EOPNOTSUPP;
1163         }
1164
1165         i->metadata_valid = true;
1166
1167         return 0;
1168 }
1169
1170 int image_name_lock(const char *name, int operation, LockFile *ret) {
1171         const char *p;
1172
1173         assert(name);
1174         assert(ret);
1175
1176         /* Locks an image name, regardless of the precise path used. */
1177
1178         if (!image_name_is_valid(name))
1179                 return -EINVAL;
1180
1181         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1182                 *ret = (LockFile) LOCK_FILE_INIT;
1183                 return 0;
1184         }
1185
1186         if (streq(name, ".host"))
1187                 return -EBUSY;
1188
1189         mkdir_p("/run/systemd/nspawn/locks", 0700);
1190         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1191
1192         return make_lock_file(p, operation, ret);
1193 }
1194
1195 bool image_name_is_valid(const char *s) {
1196         if (!filename_is_valid(s))
1197                 return false;
1198
1199         if (string_has_cc(s, NULL))
1200                 return false;
1201
1202         if (!utf8_is_valid(s))
1203                 return false;
1204
1205         /* Temporary files for atomically creating new files */
1206         if (startswith(s, ".#"))
1207                 return false;
1208
1209         return true;
1210 }
1211
1212 bool image_in_search_path(ImageClass class, const char *image) {
1213         const char *path;
1214
1215         assert(image);
1216
1217         NULSTR_FOREACH(path, image_search_path[class]) {
1218                 const char *p;
1219                 size_t k;
1220
1221                 p = path_startswith(image, path);
1222                 if (!p)
1223                         continue;
1224
1225                 /* Make sure there's a filename following */
1226                 k = strcspn(p, "/");
1227                 if (k == 0)
1228                         continue;
1229
1230                 p += k;
1231
1232                 /* Accept trailing slashes */
1233                 if (p[strspn(p, "/")] == 0)
1234                         return true;
1235
1236         }
1237
1238         return false;
1239 }
1240
1241 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1242         [IMAGE_DIRECTORY] = "directory",
1243         [IMAGE_SUBVOLUME] = "subvolume",
1244         [IMAGE_RAW] = "raw",
1245         [IMAGE_BLOCK] = "block",
1246 };
1247
1248 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);